Improve error message when future covariates are missing during backtest() call

[Jaco-Pastorius]

Is your feature request related to a current problem? Please describe.
Yes.When using backtest() with a model trained with train and val future covariates, forgetting to pass the full future covariates again at backtest time results in an error (IndexError: index 1 is out of bounds for axis 0 with size 1.).

Describe proposed solution
A better explanation or error message should be thrown as it is the case for "regular" historical forecasts.

Additional context
Error currently thrown :

IndexError: index 1 is out of bounds for axis 0 with size 1.

[dennisbader]

Thanks @Jaco-Pastorius. Do you have an example to reproduce this issue?

[AddyM]

Hi all,
I’ve started looking into this issue and able to reproduce the issue locally. . will share the findings soon.

[AddyM]

Hello ,
When calling backtest() or historical_forecasts() on models requiring future covariates (e.g., TFTModel), users encounter cryptic errors if covariates are missing or insufficient
I reproduced the issues which uncovered few more errors as below:
Errors hit :
1 .IndexError: index 1 is out of bounds for axis 0 with size 1 (when covariates too short)

2. TypeError: 'NoneType' object is not subscriptable (when covariates start too late)

3. Generic ValueError (when covariates completely missing)

Root Cause: No validation in backtest() or historical_forecasts() before passing data to _get_historical_forecasts_setup(), causing errors deep in the call stack with unclear messages.

proposed solution:
Added comprehensive validation at method entry points to provide clear, actionable error messages before processing begins.
Note , Along with Index Error mentioned in the bug this will fix a Type error and a value error .

-thanks

[AddyM]

Hello @dennisbader , raised a PR #2925 .
Added as much as possible details in the description . Please let me know if I am missing any key data . Thanks